Does the Chinese Airline Network Become More Robust Over Time?

Xiaoqian Sun Sebastian Wandelt* School of Electronic and Information Engineering Knowledge Management in Bioinformatics Beihang University Humboldt-University of Berlin Beijing, China Berlin, Germany [email protected] [email protected] (corresponding author)

Abstract— Currently, China has the fastest growing air In this study, we model the Chinese air transportation transportation market in the world. Resilience to external events system from a complex network point of view: Airports are is critical to ensure an efficient and reliable transportation of modelled as nodes; and there exists a link between two airports passengers. This study investigates the robustness of the Chinese if there is direct flight connection between both airports. In airline network under disruptions at their critical airports as well general, different airlines have their own network structures, as the evolution of the networks’ robustness from the year 2010 depending on their business model and strategical decisions. In to 2015. Among the 24 Chinese airline networks in our study, we the Chinese airline industry, there are three dominating airlines: find that the topological properties of the networks differ , China Eastern, and China Southern, whose network significantly for these airlines. Each airline has its own few structures are following the typical hub-and-spoke network dominating hubs, where the number of hubs varies among airlines. Analysis of the robustness for the 24 Chinese airline model. The rest of the Chinese airlines are significantly smaller networks show that they are quite robust against random and they often have mixed network structures. For instance, the failures, but the networks disintegrate quickly under targeted network of , the first low-cost carrier in China attacks. Evolutionary analysis of the robustness on a monthly (established in Shanghai in 2005), has been transformed from a resolution from 2010 to 2015 showed that the robustness does not star structure to a complex one with multiple hubs [2]. change significantly over time for individual airline network, Figure 1 shows an example for the airline network of although the robustness measure R values vary for different (yellow color), on top of the aggregated airlines. This shows that the ongoing efforts to increase the Chinese airline network (blue color). Chengdu Airlines was resilience should be adjusted to better meet future needs for more reliable transportation. Our work contributes to a better founded in 2004 and revenue flights were commenced starting understanding of the Chinese air transportation systems. from 2005, with CTU (Chengdu Shuangliu International Airport) as its operational hub. Based on the air traffic data for Keywords-air transportation, airline networks, robustness August 2015, Chengdu airlines served 39 airports with 50 direct flight connections, carrying approximately 0.15 million passengers. We can observe from Figure 1 that airlines often I. INTRODUCTION serve some sub-region of China only, with a strong focus on a Despite the fast spreading of the Chinese High-Speed Rail few hub airports. Note that in the Chengdu airline network, (HSR) system [11] in recent years, with the HSR tracks only airports served by this airline are highlighted with yellow summing up to 19,000 km and approximately one billion color; while in the aggregated Chinese airline network, airports passengers carried, China has the fastest growing number of air served by any Chinese airlines are modelled as nodes (yellow passengers in the world; and it has been the second largest air color and blue color). transportation market for more than a decade. In 2013, the This paper is organized as follows. Section II provides the number of air passengers carried in China is more than 352 literature on the robustness of air transportation networks. million and Beijing Capital International Airport (PEK) has Topological properties of the 24 Chinese airline networks are handled more than 77 million passengers, which is the second presented in Section III. Section IV assesses the robustness of busiest airport in the world, following ATL (Hartsfield-Jackson the 24 Chinese airline networks and presents the evolutionary Atlanta International Airport). It is predicted that the number of analysis of the network robustness with a monthly resolution aircraft operated by the Chinese airlines will be more than 4000 from 2010 to 2015. Finally, conclusions are discussed in in the year 2020. Because of the key importance roles of such Section V. hub nodes for the functionality of the Chinese air transportation system [6], it is critical to assess the system robustness under disruptions as well as the evolution of the system robustness II. LITERATURE REVIEW over the years [10]. This section provides the literature on the robustness of air transportation networks. The robustness of the worldwide

R020 The Second International Conference on Reliability Systems Engineering (ICRSE 2017) airport network with twelve different attacking strategies and III. DATA IN THIS STUDY three robustness measures was studied in [5], the goal was to An overview of the 24 Chinese airlines in this study is presented in Table 1, including the IATA codes, full names of the 24 airlines, passengers carried, number of airports (nodes), number of flight connections (links), network density, ASPL (Average Shortest Path Length, representing the average number of hops between all airports inside an airline network), and average degree (the average number of connections an airport has). There are three state-owned airlines (Air China, , and ) and they dominate the Chinese airline industry. As we can see from Table 1 (see appendix), China Southern Airlines (CZ) has the most number of passengers (3.57 Mio) and the highest number of flight connections (479) in August 2015; while China Figure 1: Example for the airline network of Chengdu airlines (yellow color), Eastern Airlines (MU) served most number of airports (124). on top of the aggregated Chinese airline network (blue color). Chengdu The density of the Chinese airline networks varies Airlines was founded in 2004 and revenue flights were commerced starting significantly from 4% to 25%; while the average number of from 2005, with CTU (Chengdu Shuangliu International Airport) as its operational hub. It can be observed that airlines often serve some sub-region hops between airports is between 2.11 to 3.66. The average of China only, with a strong focus on a few hub airports. number of connections per airport also differ for the 24 airlines, ranging from 1.8 to 8.1. identify commonalities and differences under these scenarios. It Figure 2 (see appendix) visualizes the top nine Chinese was found that degree and Bonacich-based attacks harm the passenger weighted network the most. It was also revealed that airline networks, according to the number of passengers the worldwide airport network is a redundant and resilient carried in August 2015 in our study: China Southern Airlines, network for long distance , but otherwise breaks down China Eastern Airlines, Air China, , Shenzhen completely due to removal of short insignificant connections Airlines, Xiamen Airlines, , Shandong [8]. In order to improve the resilience of worldwide air Airlines, and . In this figure, green nodes transport system under targeted node attacks, [3] proposed represent all airports inside an airline network and green lines reroute of flights within certain distances of original destination represent direct flight connections; while red nodes indicate airports, using an estimated number of stranded passengers in that these airports are hubs, who are connected to more than the giant component as a robustness metric. 45% of all airports in the respective airline network. Hubs, in general, have the function to transfer passengers from origin to The robustness of US and European air transportation networks has been intensively studied as well. [15] analyzed destination by taking into account cost discounts. In a hub- individual structures of seven US largest passenger airline based network, the transportation is significantly cheaper networks and the networks’ resilience to random node/link compared to point-to-point networks, at the price of increased deletion and targeted node deletion based on transportation times. degree/betweenness centralities. The size of giant component It is interesting to note that most of the nine airlines indeed and a relative global travel cost were used to quantify the have a few dominating hubs, from which more than 45% of all network performance under various deletion processes. [12, 13, other airports can be reached within one hop. For instance, 14] introduced the flight routes addition/deletion problem and Sichuan Airlines has two hubs (Chengdu Shuangliu used algebraic connectivity as the robustness measure to International Airport and Chongqing Jiangbei International optimize the network robustness; the Virgin America network Airport); while none of the airports meets the threshold of 45% was used as a case study. The resilience of European air connections with other airports for Tianjin Airlines. Since there transport network against random link failures was analyzed, is no unique and standard definition of hubs in the literature, in with each airline as an interdependent network [1]. It was this study we choose the threshold of 45% connections with found that the multi-layer structure strongly reduces the other airports. Note that the appearance of the hub airports for system’s resilience under disruptions. an airline depends on the threshold of 45% and the hubs identified by this definition are not necessarily the same as the Recently, the computational efficiency for the robustness operational hubs of the airlines. analysis of complex networks has drawn attention as well. Based on the combination of cheap-to-compute network IV. ROBUSTNESS OF THE CHINESE AIRLINE NETWORKS metrics, [9] proposed a new framework and implementation for estimating the robustness of a network in sub-quadratic time, The robustness of an airline network can be studied from its QRE (Quick Robustness Estimation). Experiments on twelve topological perspective: Once airports are closed due to real-world networks showed that QRE estimates the robustness disruptions (such as extreme weather conditions, mechanic better than betweenness centrality-based computation, while failures or employee strikes), how is the connectivity of the being at least one order of magnitude faster for larger networks. airline network affected? The Giant Component (GC) is one of the mostly common used measure for the network connectivity.

R020 The Second International Conference on Reliability Systems Engineering (ICRSE 2017) The GC size is the number of nodes inside the largest (MARs) could also be studied, since passengers are more likely connected component of a network. There are different to be re-accommodated to other nearby airports in case of strategies to simulate the airport closure inside an airline disruptions [7]. Since the fast-growing Chinese high-speed rail network: Random removal or targeted attacks. In this study, we system triggers increasing competition between air and rail, it use three different strategies: Random failures, degree-based would be also interesting to investigate the attacks, and betweenness-based attacks. We compute the R competition/cooperation effects between rail and air transport. values as a robustness measure for the network after an attack. The R value is defined as follows [4]: Given a network with ACKNOWLEDGMENT nodes, , where is the size of the giant component (GC size) after removing nodes. The R values This paper is supported by the National Natural Science range from 0 to 0.5; smaller R values indicate more fragility Foundation of China (Grant No. 61650110516 and Grant No. and larger R values indicate more robustness for a network 61601013).

Figure 3 (see appendix) visualizes the robustness curves for REFERENCES the 24 Chinese airline networks in this study, induced by different attacks to the network: Random failures (red), degree- [1] A. Cardillo, M. Zanin, J. Gomez-Gardenes, M. Romance, A. J. Garcia del based attacks (blue), and betweenness-based attacks (green). Amo, and S. Boccaletti. Modeling the multi-layer nature of the european The R values under the betweenness-based attacks are shown air transport network: Resilience and passengers re-scheduling under as well. We can observe that most airlines are rather robust random failures. The European Physical Journal Special Topics, 215(1), against random airport failures, but they are quite fragile under pp. 23–33, 2013. targeted attacks. It is also interesting to observe that the [2] Y. Jiang, B. Yao, L. Wang, T. Feng, and L. Kong. Evolution trends of the attacking strategies based on degree and betweenness behave network structure of spring airlines in china: A temporal and spatial analysis. Journal of Air Transport Management, 60, pp. 18 – 30, 2017. quite similar in the Chinese airline networks, indicating that it [3] V. H. Louzada, N. A. Araújo, T. Verma, F. Daolio, H. J. Herrmann, and only takes a few nodes to break down the connectivity of the M. Tomassini. Critical cooperation range to improve spatial network network, once sufficient knowledge about the network (such as robustness. PloS one, 10(3), pp. e0118635, 2015. the structural properties) is obtained. Note that the [4] C. M. Schneider, A. A. Moreira, J. S. Andrade, S. Havlin, and H. J. effectiveness of random failures and targeted attacks become Herrmann. Mitigation of malicious attacks on networks. Proceedings of similar for very small networks, such as Joy Airlines in this the National Academy of Sciences, 108(10), pp. 3838–3841, 2011. study. [5] X. Sun, V. Gollnick, and S. Wandelt. Robustness analysis metrics for worldwide airport network: A comprehensive study. Chinese Journal of Figure 4 (see appendix) shows the evolution of the Aeronautics, 30(2), pp. 500–512, 2017. robustness for the 24 airline networks in China, these networks [6] X. Sun, S. Wandelt, and F. Linke. On the topology of air navigation route are built on a monthly resolution from 2010 to 2015. The systems. Proceedings of the Institution of Civil Engineers-Transport, 170 robustness is measured by the R values. It can be observed that (1), pp. 46-59, 2017. for most airline networks, although the R values vary for [7] X. Sun, S. Wandelt, M. Hansen, and A. Li. Multiple airport regions based different airline networks, the robustness does not change on inter-airport temporal distances. Transportation Research Part E: Logistics and Transportation Review, 101, pp. 84 – 98, 2017. significantly over the time for individual airline networks. Note [8] T. Verma, N. A. Araújo, and H. J. Herrmann. Revealing the structure of that because of the lack of data, the R values for the following the world airline network. Scientific reports, , pp. 4, 2014. three airlines (bottom of the chart): (commenced [9] S. Wandelt, X. Sun, M. Zanin, and S. Havlin. QRE: Quick Robustness operations since 2011), Joy Airlines (commenced operations Estimation for large complex networks. Future Generation Computer since 2009), and JSC Starline (also named as , Systems, in press, 2017, doi:10.1016/j.future.2017.02.018. commenced operations since 2006), are discontinuous. [10] S. Wandelt, X. Sun, and J. Zhang. Evolution of domestic airport networks: a review and comparative analysis. Transportmetrica B: Transport Dynamics, in press, 2017, V. CONCLUSIONS doi:10.1080/21680566.2017.1301274. [11] S. Wandelt, Z. Wang, and X. Sun. Worldwide railway skeleton network: This study investigated the robustness of the Chinese airline Extraction methodology and preliminary analysis. IEEE Transactions on network under disruptions at their critical airports as well as the Intelligent Transportation Systems, in press, 2016, doi: 10.1109/TITS.2016.2632998. evolution of their robustness from 2010 to 2015. Among the 24 [12] P. Wei, L. Chen, and D. Sun. Algebraic connectivity maximization of an Chinese airline networks in our study, the topological air transportation network: The flight routes addition/deletion problem. properties of the networks differ significantly for these airlines. Transportation Research Part E: Logistics and Transportation Review, Each airline has its own few dominating hubs. Analysis of the 61(0), pp. 13–27, 2014. robustness curves for the 24 Chinese airline networks showed [13] P. Wei, G. Spiers, and D. Sun. Algebraic connectivity maximization for that they are quite robust against random failures, but the air transportation networks. Intelligent Transportation Systems, IEEE Transactions on, 15(2), pp. 685–698, April 2014. networks disintegrate quickly under targeted attacks. [14] P. Wei and D. Sun. Weighted algebraic connectivity: An application to Evolutionary analysis of the robustness on a monthly resolution airport transportation network. In 18th IFAC World Congress Milano from 2010 to 2015 showed that the robustness does not change (Italy), 2011. significantly over time for individual airline networks, although [15] D. R. Wuellner, S. Roy, and R. M. D’Souza. Resilience and rewiring of the robustness measure R values vary for different airlines. the passenger airline networks in the united states. Physical Review E, Future work could take into account passenger flows on top of 82(5), pp. 056101, 2010. the topological network [5]. The multiple-airport regions

R020 The Second International Conference on Reliability Systems Engineering (ICRSE 2017) APPENDIX

Table 1: List of airlines in our study, ordered by the number of passengers carried (descending), data for August 2015. The topological parameters differ significantly for these airlines. ASPL=Average shortest path length. The maximum values for each column are highlighted in bold.

ID IA Name Passengers Node Links Density ASPL Avg. degree TA s 1 CZ CHINA SOUTHERN 3.57 Mio 118 479 6.94% 2.38 8.1 AIRLINES 2 M CHINA EASTERN 2.94 Mio 124 406 5.32% 2.45 6.5 U AIRLINES 3 CA AIR CHINA 1.83 Mio 102 251 4.87% 2.13 4.9 LIMITED 4 HU HAINAN AIRLINES 1.19 Mio 59 196 11.46% 2.23 6.6 COMPANY LIMITE... 5 ZH SHENZHEN 1.17 Mio 61 187 10.22% 2.15 6.1 AIRLINES 6 MF XIAMEN AIRLINES 1.02 Mio 57 178 11.15% 2.17 6.2 7 3U SICHUAN AIRLINES 1.01 Mio 76 205 7.19% 2.36 5.4 CO. LTD. 8 SC SHANDONG 0.74 Mio 55 140 9.43% 2.33 5.1 AIRLINES 9 GS TIANJIN AIRLINES 0.53 Mio 80 152 4.81% 2.89 3.8 CO. LTD 10 FM SHANGHAI 0.50 Mio 62 80 4.23% 2.72 2.6 AIRLINES CO. LTD. 11 9C SPRING AIRLINES 0.49 Mio 46 78 7.54% 2.49 3.4 LIMITED CORPOR... 12 JD BEIJING CAPITAL 0.45 Mio 52 122 9.20% 2.55 4.7 AIRLINES CO. L... 13 HO JUNEYAO 0.36 Mio 47 67 6.20% 2.41 2.9 AIRLINES CO. LTD. 14 8L CO. 0.31 Mio 43 65 7.20% 2.50 3.0 LTD. 15 PN CHINA 0.21 Mio 35 46 7.73% 2.36 2.6 CO. LTD. 16 BK 0.19 Mio 44 60 6.34% 2.85 2.7 COMPANY LIMITED 17 EU CHENGDU 0.15 Mio 39 50 6.75% 2.50 2.6 AIRLINES 18 KY KUNMING 0.12 Mio 26 29 8.92% 2.18 2.2 AIRLINES 19 OQ CHONGQING 0.10 Mio 24 24 8.70% 2.06 2.0 AIRLINES CO. LTD 20 G5 CHINA EXPRESS 0.09 Mio 51 55 4.31% 3.66 2.2 AIRLINES 21 NS 0.09 Mio 24 28 10.14% 2.38 2.3 CO. LTD. 22 TV TIBET AIRLINES 0.06 Mio 18 21 13.73% 2.40 2.3 CORPORATION LIM... 23 JR JOY AIR 0.05 Mio 17 17 12.50% 2.87 2.0 24 DZ JSC STARLINE.KZ 0.03 Mio 8 7 25.00% 2.11 1.8

R020 The Second International Conference on Reliability Systems Engineering (ICRSE 2017)

Figure 2: Visualization of the top nine Chinese airline networks, according to the number of passengers carried in August 2015 (the ranking in Table 1. In each sub-figure, blue nodes represent all airports inside an airline network and blue lines represent direct flight connections; while yellow nodes indicate that these airports are hubs, who are connected to more than 45% of all airports in the airline network.

Figure 3: Visualization of robustness curves for the 24 Chinese airline networks in this study, induced by different attacks to the network: Random failures (red, dotted), degree-based attacks (blue, line), and betweenness-based attacks (green, dashed). The R values under betweenness-based attacks are shown as well.

R020 The Second International Conference on Reliability Systems Engineering (ICRSE 2017)

Figure 4: Evolution of the robustness on a monthly resolution for the 24 airline networks in China from 2010 to 2015. The robustness is measured by the R values. For most airline networks, although the R values vary for different airline networks, the robustness does not change significantly over the time for individual airline networks.

R020 The Second International Conference on Reliability Systems Engineering (ICRSE 2017)